For 2025 the grid carbon intensity in Texas was 363gCO2eq/kWh, while it was 22gCO2eq/kWh in Sweden, so 16,5x more in Texas. Right now it's 427 in Texas vs. 50 in Finland ie. 8,54x more, like in the article (yes I looked the wrong country at first).
The reason is half of Texas electricity production is still fossil fuels (gas & coal) while mostly hydro, nuclear and wind in Sweden. So it's not really about ai, the models or the hardware. Fossil fuels make CO2, that's it.
The comments section reminds me of The Dead Great-Grandfather Test
[Science fiction writer Bruce] Sterling was using the term specifically to describe the hairshirt green lifestyle that is driven by eco-anxieties. For hairshirt-green types, life is all about saving water, recycling, composting, reducing eco-footprints and various other behaviors marked by a kind of fearful, non-generative retreat from living. Permanent existential hibernation.
Sterling's rule of thumb for spotting acting-dead behaviors is a great one: if it's something your dead great-grandfather can do better than you, it's a case of acting dead. Your dead great-grandfather uses no water or plastic, and is actually recycling himself as we speak, not just his possessions. Try and top that.
But acting dead goes beyond hairshirt-green behaviors. While spartan frugality is a virtue, when it becomes the entire purpose of your life, there's a problem. For a portion of the dying American middle class, frugality has turned into a life purpose.
Is getting in touch with the earth/spending time outside, appreciating the value in things and protecting nature really 'Acting Dead'?
I would say 'Acting Dead' is actually sitting at home ordering on Amazon/Temu, watching brainless TikTok/Instagram/Netflix/YouTube videos, and talking to a transformer model all day.
This is a fallacy. People 100 years like your great grandfather ago solved problems they had 100 years ago. People today solve problems they have today.
This is about a dead grandfather as he is decomposing in the ground. Not about his life which might have been more energy-intensive than that of many moderns.
Not sure about your great-grandfather, but mine didn't own a car, didn't travel on vacation by airplane, and, while he may have used a stove to heat his house, he probably only used it sparingly, because coal was expensive. So I doubt his life was more energy-intensive than mine.
It's a really dumb test. You can apply it to argue against reducing or stopping literally anything. My dead great grandfather uses no fentanyl and commits no murders. If we're not doing those, we're as good as dead!
Basically "Don't live beyond our collective means" = "Die".
Are you buying the latest iPhone every year? If not, you might as well be dead.
I wish people would realize that life is not about stuff. Living within your means, being frugal and focusing on what really matters (time with family, experiencing nature, thinking interesting thoughts) is not at all "permanent existential hibernation".
I resent the "The Dead Great-Grandfather Test" and similar "growth"-at-all-costs ideologies. Its implications are disgusting.
Yeah, the average Swede emitted 3.43 tons of CO2 in 2023, which is less than a quarter of what an average US citizen emitted. Does that mean that a US citizen is four times more "alive" than a Swede?
The reasoning effort knob is becoming a really important variable regarding how much impact AI models will have on energy consumption. Medium and max can differ by as much as two orders of magnitude in their token consumption. We also have to consider the amplification of compute demand that occurs by way of the tool calls themselves. Tokens are mostly a constant cost thing, but how they are interpreted can mean the difference between a 5 minute cpu bound operation and a 50uS cached database lookup. At some point, the CPU is going to become the bottleneck again.
Great writeup! I've done some work on Codecarbon [1] in the past, it's a pretty standard tool for measuring ML/AI workload emissions, it would be awesome to have your findings contributed to that project as well.
And eating a single burger results in the equivalent Co2 emissions of 17,000 requests in Sweden and 2,000 requests in Texas.
Reducing/stopping meat consumption and replacing it with healthy plant-based tofu, seitan, tempeh etc has a far bigger impact (and as a bonus, living beings don't have to suffer)
That's actually kind of a scary statistic... I probably eat a burgur at most every month (beef maybe once a week at most) but probably make more than 2000 requests a day.
Your comment makes me really contextualize my AI usage and want to reduce it for ecological reasons.
0.5 Pwh/yr - worldwide data center energy
4-5 PWh/yr - meat total farm to table
166 PWh/yr - worldwide total energy production
the true bottleneck and cost we have is energy, carbon is a polite proxy
if you can meaningfully cut down the energetic cost, that's very valuable either way
maybe tomorrow you'll want to spin up a GPU on the cloud to briefly run a model, pick a data center:
DC #1: EU Efficient, but has ~100ms additional latency
DC #2: TX Runs on jet engine turbines, neighbors complain of noise and air pollution
env impact is something worth considering regardless of industry
and life of subsistence farming in the countryside costs two orders of magnitude less of carbon sin that life in the city, where everything the enlightened cosmopolitan consumes is delivered by carbon-chugging trucks, freighters, and planes from hundreds and thousands miles away.
but the line is clearly drawn at performative sacrifice, not substantial.
I've seen people who hate AI instantly become defensive as soon as someone brings up the water usage for a kg of beef. It's selective outrage.
I don't like whataboutism, but this isn't whataboutism because the meat industry requires way more water per capita like you pointed out. You have to create products twice, once for feed and once to raise the cows.
The meat industry helps keep humans alive however, whereas the AI industry does not. I think the argument between the two needs to factor in the actual reason for the emissions.
Eating meat is ecologically unsustainable but it has an actual reason for it to exist.
Eating meat is not necessary by any means to 'be alive' in fact many studies show that a plant based diet is better for longevity. So, in the same way that AI is not 'necessary' to do a job, it's not necessary to eat meat. It's a matter of taste and societal conformity/norms.
Because some causes are significantly worse than others. If you want to have a big impact, you start with the big levers.
I'll give you an even better example than meat: golf courses in the US consume 2 to 100 times as much water as US datacenters depending on how you calculate.
How many golf-course-water-saving initiatives have you seen? Have they even tried anything at all?
Have a look at a graph of meat consumption in the US. We are also adding to that. That's my point, we don't need to add to meat consumption, we can make reductions.
It might not help the cause, but ultimately it's an uncomfortable truth. It's easy to dismiss a fact as 'moralising'.
It's simply a fact that animals suffer when they are factory farmed and then killed. It's a fact that they are living sentient beings. You can be totally okay with that, but the reality is that most people are only okay with it because it doesn't happen right in front of them.
I used to eat meat, btw. And I also used to attack people for 'moralising'. I now realise this was my brain trying to defend itself; to try to make sense of why I would pet a dog but eat a pig (everyone else is doing it, this person is just different and wrong, go away bad vegan person!)
The relative resource consumption of meat vs. plants is well established. Chicken is much more resource efficient than beef if you want to eat meat but consume less resources, and mussels are even more efficient.
That part I'm well aware of, I wanna know how the parent reached the figure 17000. What does count as a request even nowadays? When I tell an agent to make the button blue, how many requests is that? And how about when I tell it to throw together a whole website? Or rewrite the codebase in Rust?
“A typical request to the open model Kimi K3 emits about 0.4 grams of CO₂ in Sweden, 1.6 grams in Ireland, 2.5 grams in Germany and 3.2 grams in Texas, on identical hardware — 8.5 times more in Texas than in Sweden.”
0.4 x 17,000 = 6,800g, a fairly typical estimate for the co2 emissions of a burger.
Thanks. My next question would be what's a typical request. The article defines it as "production-typical request of 5,000 input and 387 output tokens, weighted by a typical day's traffic" for the Kimi K3 model. Not sure how that translates to my scale of the blue button to RIIR.
Most people in the West eat too much meat. Red meat in particular is a group 2 carcinogen according to the WHO, so the body isn't meant to consume a lot of it. Yet many people eat meat every day, because "It's not a complete meal without it" or "I somehow need 10 times more protein than people a millenium ago while claiming to eat like them".
too many things can happen in the ai provider's backend which won't be observable from the outside. for starters how will you even know what hardware and inference backend is being used by RandomAiProvider3284.com? you can't. i looked at your berget-ai/co2-calculator repo and you're expected to provide this yourself. it also uses "measuredResponseTimeSeconds" as a metric even though it's impossible to know how much of that time was spent rate limiting the user during hours of high activity
> for starters how will you even know what hardware and inference backend is being used by RandomAiProvider3284.com? you can't.
From the article:
A typical request to the open model Kimi K3 emits about 0.4 grams of CO₂ in Sweden, 1.6 grams in Ireland, 2.5 grams in Germany and 3.2 grams in Texas, on identical hardware — 8.5 times more in Texas than in Sweden.
-> So the comparison is based on the assumption that identical hardware is used.
Also from the article:
Emissions estimates are based on the current average emissions intensity of each regional electricity grid, which has been the norm in greenhouse gas accounting.
So it's just an incredibly stupid way to compare the power grids in Texas vs Sweden. It's not even accurate because the marginal watt during the day looks very different than the overall grid average.
Worry not, AI people! Greenhouse gas exemptions will be made for the future of work, maths and art, the salvation of the humankind, the genie that is out of the bottle, etc, if only at the expense of taxing your and your dog's fart.
One of the reasons to build them near populated areas is that it makes it easier to repurpose waste heat to, for example, heat houses in colder climates.
Probably so they have no problems getting people who work there and provide technical services. I mean that seriously without any sarcasm or tongue-in-cheek, it seems the most likely explanation to me.
This is why people in left wing areas need to oppose datacenters bans. Regulate them. Require green energy investments that more than offset their load. Use some of that flowing AI money for the betterment of society.
You don't need to specifically regulate green investments for datacenter. Just introduce CO2 emission tax and the industry will find efficient ways to reduce CO2 emissions.
I think "people in left wing areas" is weird phrasing. There is nothing inherent about an area that makes it left or right wing. "People who live under governments who are responsive to collective issues such as climate and energy prices" would make more sense.
Some of it is useful, and from an environmental perspective, if it eliminates some other activity (e.g. team of workers driving to/from work and their office heating/cooling) then it can even be a net positive.
I would agree that there should be some tariffed token usage based on CO2, efficient datacentres pay little, polluting ones pay a lot - would help even the field and export good practice.
Datacenter wouldn't stop, and there would be some country or area which allows it. Unless they regulate the entire world, we should focus on lowering CO2 globally, and that means allowing datacenter in your neighour if your energy is better.
We can pretend that international trade and energy price arbitrage doesn't exist to make us feel nice and happy; but the reality is a Watt spent in Sweden increases the marginal cost of energy by the same amount as a Watt spent in Texas. These things are tied together by foreign exchange, imported energy intensive goods, petroleum products and directly imported power. It's time we stop being naive about total energy use.
Sweden is not grid connected to Texas, well, nor is also any other grid in the USA actually. Smelting was placed near cheap hydro resources (and why Boeing built planes in Seattle) for good reasons. Just replace cheap watts with clean watts.
Because Sweden's clean energy capacity is fixed. Which means a "clean" Watt spent on inference is taking away a "clean" Watt that would have been spent on heating or steel. Which means Sweden needs to buy energy or energy intensive goods on the global market, which raises prices, which means Texas is being paid.
This is a weird assumption to make. Why would you assume they wouldn't just build out more capacity as demand increases?
They certainly aren't buying electricity from Texas, this feels like a tortured argument you're just making to retroactively support whatever opinions you have on the pros or cons of clean energy.
There's only so much hydro power to build out, and nuclear takes forever. Solar only works in summer that far north, and without insane batteries wind power just becomes greenwashed natgas.
I'm a bit confused. For one I am not sure why this must hold for something as difficult to transfer as energy, and I am not sure why it matters. Not all energy is the same, hence the title.
The reason is half of Texas electricity production is still fossil fuels (gas & coal) while mostly hydro, nuclear and wind in Sweden. So it's not really about ai, the models or the hardware. Fossil fuels make CO2, that's it.
[Science fiction writer Bruce] Sterling was using the term specifically to describe the hairshirt green lifestyle that is driven by eco-anxieties. For hairshirt-green types, life is all about saving water, recycling, composting, reducing eco-footprints and various other behaviors marked by a kind of fearful, non-generative retreat from living. Permanent existential hibernation.
Sterling's rule of thumb for spotting acting-dead behaviors is a great one: if it's something your dead great-grandfather can do better than you, it's a case of acting dead. Your dead great-grandfather uses no water or plastic, and is actually recycling himself as we speak, not just his possessions. Try and top that.
But acting dead goes beyond hairshirt-green behaviors. While spartan frugality is a virtue, when it becomes the entire purpose of your life, there's a problem. For a portion of the dying American middle class, frugality has turned into a life purpose.
https://ribbonfarm.com/2011/12/08/acting-dead-trading-up-and...
I would say 'Acting Dead' is actually sitting at home ordering on Amazon/Temu, watching brainless TikTok/Instagram/Netflix/YouTube videos, and talking to a transformer model all day.
Are you buying the latest iPhone every year? If not, you might as well be dead.
I wish people would realize that life is not about stuff. Living within your means, being frugal and focusing on what really matters (time with family, experiencing nature, thinking interesting thoughts) is not at all "permanent existential hibernation".
I resent the "The Dead Great-Grandfather Test" and similar "growth"-at-all-costs ideologies. Its implications are disgusting.
[1] https://github.com/mlco2/codecarbon
Reducing/stopping meat consumption and replacing it with healthy plant-based tofu, seitan, tempeh etc has a far bigger impact (and as a bonus, living beings don't have to suffer)
Your comment makes me really contextualize my AI usage and want to reduce it for ecological reasons.
maybe tomorrow you'll want to spin up a GPU on the cloud to briefly run a model, pick a data center:
env impact is something worth considering regardless of industrybut the line is clearly drawn at performative sacrifice, not substantial.
That will put some sweat on some brows :)
I don't like whataboutism, but this isn't whataboutism because the meat industry requires way more water per capita like you pointed out. You have to create products twice, once for feed and once to raise the cows.
Eating meat is ecologically unsustainable but it has an actual reason for it to exist.
I'll give you an even better example than meat: golf courses in the US consume 2 to 100 times as much water as US datacenters depending on how you calculate.
How many golf-course-water-saving initiatives have you seen? Have they even tried anything at all?
What benefit is the AI summary on search giving me Vs the actual search for a start.
if you dont leave the page they get more ad-money
Flying more is a thing we are choosing to do, we could just not do that.
The "only the biggest thing matters" is a faulty logic. If you truly care about the environment then we need to improve things across the board.
And moralizing about living beings doesn't help your cause here.
It's simply a fact that animals suffer when they are factory farmed and then killed. It's a fact that they are living sentient beings. You can be totally okay with that, but the reality is that most people are only okay with it because it doesn't happen right in front of them.
I used to eat meat, btw. And I also used to attack people for 'moralising'. I now realise this was my brain trying to defend itself; to try to make sense of why I would pet a dog but eat a pig (everyone else is doing it, this person is just different and wrong, go away bad vegan person!)
https://en.wikipedia.org/wiki/Environmental_impacts_of_anima...
“A typical request to the open model Kimi K3 emits about 0.4 grams of CO₂ in Sweden, 1.6 grams in Ireland, 2.5 grams in Germany and 3.2 grams in Texas, on identical hardware — 8.5 times more in Texas than in Sweden.”
0.4 x 17,000 = 6,800g, a fairly typical estimate for the co2 emissions of a burger.
From the article:
A typical request to the open model Kimi K3 emits about 0.4 grams of CO₂ in Sweden, 1.6 grams in Ireland, 2.5 grams in Germany and 3.2 grams in Texas, on identical hardware — 8.5 times more in Texas than in Sweden.
-> So the comparison is based on the assumption that identical hardware is used.
Also from the article:
Emissions estimates are based on the current average emissions intensity of each regional electricity grid, which has been the norm in greenhouse gas accounting.
Power connections.
How much of the AI generated content is actually useful, how much useless, how much harmful?
I would agree that there should be some tariffed token usage based on CO2, efficient datacentres pay little, polluting ones pay a lot - would help even the field and export good practice.
That content is coming either way.
And if you require them to over offset their load to the grid, it's literally fossil fuels burned over all.
This is a weird assumption to make. Why would you assume they wouldn't just build out more capacity as demand increases?
They certainly aren't buying electricity from Texas, this feels like a tortured argument you're just making to retroactively support whatever opinions you have on the pros or cons of clean energy.
My opinion that reality should take precedent over feel good lies?
By what?
You're treating electricity as a globally-tradable good. It's not.
1. Hydropower 40%
2. Nuclear 27%
3. Wind 22%
https://ourworldindata.org/profile/energy/sweden#what-source...